2009-05-20 21:30:49 +07:00
|
|
|
/*
|
|
|
|
* kvm eventfd support - use eventfd objects to signal various KVM events
|
|
|
|
*
|
|
|
|
* Copyright 2009 Novell. All Rights Reserved.
|
|
|
|
*
|
|
|
|
* Author:
|
|
|
|
* Gregory Haskins <ghaskins@novell.com>
|
|
|
|
*
|
|
|
|
* This file is free software; you can redistribute it and/or modify
|
|
|
|
* it under the terms of version 2 of the GNU General Public License
|
|
|
|
* as published by the Free Software Foundation.
|
|
|
|
*
|
|
|
|
* This program is distributed in the hope that it will be useful,
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
* GNU General Public License for more details.
|
|
|
|
*
|
|
|
|
* You should have received a copy of the GNU General Public License
|
|
|
|
* along with this program; if not, write to the Free Software Foundation,
|
|
|
|
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/kvm_host.h>
|
KVM: add ioeventfd support
ioeventfd is a mechanism to register PIO/MMIO regions to trigger an eventfd
signal when written to by a guest. Host userspace can register any
arbitrary IO address with a corresponding eventfd and then pass the eventfd
to a specific end-point of interest for handling.
Normal IO requires a blocking round-trip since the operation may cause
side-effects in the emulated model or may return data to the caller.
Therefore, an IO in KVM traps from the guest to the host, causes a VMX/SVM
"heavy-weight" exit back to userspace, and is ultimately serviced by qemu's
device model synchronously before returning control back to the vcpu.
However, there is a subclass of IO which acts purely as a trigger for
other IO (such as to kick off an out-of-band DMA request, etc). For these
patterns, the synchronous call is particularly expensive since we really
only want to simply get our notification transmitted asychronously and
return as quickly as possible. All the sychronous infrastructure to ensure
proper data-dependencies are met in the normal IO case are just unecessary
overhead for signalling. This adds additional computational load on the
system, as well as latency to the signalling path.
Therefore, we provide a mechanism for registration of an in-kernel trigger
point that allows the VCPU to only require a very brief, lightweight
exit just long enough to signal an eventfd. This also means that any
clients compatible with the eventfd interface (which includes userspace
and kernelspace equally well) can now register to be notified. The end
result should be a more flexible and higher performance notification API
for the backend KVM hypervisor and perhipheral components.
To test this theory, we built a test-harness called "doorbell". This
module has a function called "doorbell_ring()" which simply increments a
counter for each time the doorbell is signaled. It supports signalling
from either an eventfd, or an ioctl().
We then wired up two paths to the doorbell: One via QEMU via a registered
io region and through the doorbell ioctl(). The other is direct via
ioeventfd.
You can download this test harness here:
ftp://ftp.novell.com/dev/ghaskins/doorbell.tar.bz2
The measured results are as follows:
qemu-mmio: 110000 iops, 9.09us rtt
ioeventfd-mmio: 200100 iops, 5.00us rtt
ioeventfd-pio: 367300 iops, 2.72us rtt
I didn't measure qemu-pio, because I have to figure out how to register a
PIO region with qemu's device model, and I got lazy. However, for now we
can extrapolate based on the data from the NULLIO runs of +2.56us for MMIO,
and -350ns for HC, we get:
qemu-pio: 153139 iops, 6.53us rtt
ioeventfd-hc: 412585 iops, 2.37us rtt
these are just for fun, for now, until I can gather more data.
Here is a graph for your convenience:
http://developer.novell.com/wiki/images/7/76/Iofd-chart.png
The conclusion to draw is that we save about 4us by skipping the userspace
hop.
--------------------
Signed-off-by: Gregory Haskins <ghaskins@novell.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
2009-07-08 04:08:49 +07:00
|
|
|
#include <linux/kvm.h>
|
2009-05-20 21:30:49 +07:00
|
|
|
#include <linux/workqueue.h>
|
|
|
|
#include <linux/syscalls.h>
|
|
|
|
#include <linux/wait.h>
|
|
|
|
#include <linux/poll.h>
|
|
|
|
#include <linux/file.h>
|
|
|
|
#include <linux/list.h>
|
|
|
|
#include <linux/eventfd.h>
|
KVM: add ioeventfd support
ioeventfd is a mechanism to register PIO/MMIO regions to trigger an eventfd
signal when written to by a guest. Host userspace can register any
arbitrary IO address with a corresponding eventfd and then pass the eventfd
to a specific end-point of interest for handling.
Normal IO requires a blocking round-trip since the operation may cause
side-effects in the emulated model or may return data to the caller.
Therefore, an IO in KVM traps from the guest to the host, causes a VMX/SVM
"heavy-weight" exit back to userspace, and is ultimately serviced by qemu's
device model synchronously before returning control back to the vcpu.
However, there is a subclass of IO which acts purely as a trigger for
other IO (such as to kick off an out-of-band DMA request, etc). For these
patterns, the synchronous call is particularly expensive since we really
only want to simply get our notification transmitted asychronously and
return as quickly as possible. All the sychronous infrastructure to ensure
proper data-dependencies are met in the normal IO case are just unecessary
overhead for signalling. This adds additional computational load on the
system, as well as latency to the signalling path.
Therefore, we provide a mechanism for registration of an in-kernel trigger
point that allows the VCPU to only require a very brief, lightweight
exit just long enough to signal an eventfd. This also means that any
clients compatible with the eventfd interface (which includes userspace
and kernelspace equally well) can now register to be notified. The end
result should be a more flexible and higher performance notification API
for the backend KVM hypervisor and perhipheral components.
To test this theory, we built a test-harness called "doorbell". This
module has a function called "doorbell_ring()" which simply increments a
counter for each time the doorbell is signaled. It supports signalling
from either an eventfd, or an ioctl().
We then wired up two paths to the doorbell: One via QEMU via a registered
io region and through the doorbell ioctl(). The other is direct via
ioeventfd.
You can download this test harness here:
ftp://ftp.novell.com/dev/ghaskins/doorbell.tar.bz2
The measured results are as follows:
qemu-mmio: 110000 iops, 9.09us rtt
ioeventfd-mmio: 200100 iops, 5.00us rtt
ioeventfd-pio: 367300 iops, 2.72us rtt
I didn't measure qemu-pio, because I have to figure out how to register a
PIO region with qemu's device model, and I got lazy. However, for now we
can extrapolate based on the data from the NULLIO runs of +2.56us for MMIO,
and -350ns for HC, we get:
qemu-pio: 153139 iops, 6.53us rtt
ioeventfd-hc: 412585 iops, 2.37us rtt
these are just for fun, for now, until I can gather more data.
Here is a graph for your convenience:
http://developer.novell.com/wiki/images/7/76/Iofd-chart.png
The conclusion to draw is that we save about 4us by skipping the userspace
hop.
--------------------
Signed-off-by: Gregory Haskins <ghaskins@novell.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
2009-07-08 04:08:49 +07:00
|
|
|
#include <linux/kernel.h>
|
|
|
|
|
|
|
|
#include "iodev.h"
|
2009-05-20 21:30:49 +07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* --------------------------------------------------------------------
|
|
|
|
* irqfd: Allows an fd to be used to inject an interrupt to the guest
|
|
|
|
*
|
|
|
|
* Credit goes to Avi Kivity for the original idea.
|
|
|
|
* --------------------------------------------------------------------
|
|
|
|
*/
|
|
|
|
|
|
|
|
struct _irqfd {
|
|
|
|
struct kvm *kvm;
|
|
|
|
struct eventfd_ctx *eventfd;
|
|
|
|
int gsi;
|
|
|
|
struct list_head list;
|
|
|
|
poll_table pt;
|
|
|
|
wait_queue_head_t *wqh;
|
|
|
|
wait_queue_t wait;
|
|
|
|
struct work_struct inject;
|
|
|
|
struct work_struct shutdown;
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct workqueue_struct *irqfd_cleanup_wq;
|
|
|
|
|
|
|
|
static void
|
|
|
|
irqfd_inject(struct work_struct *work)
|
|
|
|
{
|
|
|
|
struct _irqfd *irqfd = container_of(work, struct _irqfd, inject);
|
|
|
|
struct kvm *kvm = irqfd->kvm;
|
|
|
|
|
|
|
|
kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 1);
|
|
|
|
kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Race-free decouple logic (ordering is critical)
|
|
|
|
*/
|
|
|
|
static void
|
|
|
|
irqfd_shutdown(struct work_struct *work)
|
|
|
|
{
|
|
|
|
struct _irqfd *irqfd = container_of(work, struct _irqfd, shutdown);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Synchronize with the wait-queue and unhook ourselves to prevent
|
|
|
|
* further events.
|
|
|
|
*/
|
|
|
|
remove_wait_queue(irqfd->wqh, &irqfd->wait);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We know no new events will be scheduled at this point, so block
|
|
|
|
* until all previously outstanding events have completed
|
|
|
|
*/
|
|
|
|
flush_work(&irqfd->inject);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* It is now safe to release the object's resources
|
|
|
|
*/
|
|
|
|
eventfd_ctx_put(irqfd->eventfd);
|
|
|
|
kfree(irqfd);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* assumes kvm->irqfds.lock is held */
|
|
|
|
static bool
|
|
|
|
irqfd_is_active(struct _irqfd *irqfd)
|
|
|
|
{
|
|
|
|
return list_empty(&irqfd->list) ? false : true;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Mark the irqfd as inactive and schedule it for removal
|
|
|
|
*
|
|
|
|
* assumes kvm->irqfds.lock is held
|
|
|
|
*/
|
|
|
|
static void
|
|
|
|
irqfd_deactivate(struct _irqfd *irqfd)
|
|
|
|
{
|
|
|
|
BUG_ON(!irqfd_is_active(irqfd));
|
|
|
|
|
|
|
|
list_del_init(&irqfd->list);
|
|
|
|
|
|
|
|
queue_work(irqfd_cleanup_wq, &irqfd->shutdown);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Called with wqh->lock held and interrupts disabled
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
irqfd_wakeup(wait_queue_t *wait, unsigned mode, int sync, void *key)
|
|
|
|
{
|
|
|
|
struct _irqfd *irqfd = container_of(wait, struct _irqfd, wait);
|
|
|
|
unsigned long flags = (unsigned long)key;
|
|
|
|
|
|
|
|
if (flags & POLLIN)
|
|
|
|
/* An event has been signaled, inject an interrupt */
|
|
|
|
schedule_work(&irqfd->inject);
|
|
|
|
|
|
|
|
if (flags & POLLHUP) {
|
|
|
|
/* The eventfd is closing, detach from KVM */
|
|
|
|
struct kvm *kvm = irqfd->kvm;
|
|
|
|
unsigned long flags;
|
|
|
|
|
|
|
|
spin_lock_irqsave(&kvm->irqfds.lock, flags);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We must check if someone deactivated the irqfd before
|
|
|
|
* we could acquire the irqfds.lock since the item is
|
|
|
|
* deactivated from the KVM side before it is unhooked from
|
|
|
|
* the wait-queue. If it is already deactivated, we can
|
|
|
|
* simply return knowing the other side will cleanup for us.
|
|
|
|
* We cannot race against the irqfd going away since the
|
|
|
|
* other side is required to acquire wqh->lock, which we hold
|
|
|
|
*/
|
|
|
|
if (irqfd_is_active(irqfd))
|
|
|
|
irqfd_deactivate(irqfd);
|
|
|
|
|
|
|
|
spin_unlock_irqrestore(&kvm->irqfds.lock, flags);
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
irqfd_ptable_queue_proc(struct file *file, wait_queue_head_t *wqh,
|
|
|
|
poll_table *pt)
|
|
|
|
{
|
|
|
|
struct _irqfd *irqfd = container_of(pt, struct _irqfd, pt);
|
|
|
|
|
|
|
|
irqfd->wqh = wqh;
|
|
|
|
add_wait_queue(wqh, &irqfd->wait);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
kvm_irqfd_assign(struct kvm *kvm, int fd, int gsi)
|
|
|
|
{
|
|
|
|
struct _irqfd *irqfd;
|
|
|
|
struct file *file = NULL;
|
|
|
|
struct eventfd_ctx *eventfd = NULL;
|
|
|
|
int ret;
|
|
|
|
unsigned int events;
|
|
|
|
|
|
|
|
irqfd = kzalloc(sizeof(*irqfd), GFP_KERNEL);
|
|
|
|
if (!irqfd)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
irqfd->kvm = kvm;
|
|
|
|
irqfd->gsi = gsi;
|
|
|
|
INIT_LIST_HEAD(&irqfd->list);
|
|
|
|
INIT_WORK(&irqfd->inject, irqfd_inject);
|
|
|
|
INIT_WORK(&irqfd->shutdown, irqfd_shutdown);
|
|
|
|
|
|
|
|
file = eventfd_fget(fd);
|
|
|
|
if (IS_ERR(file)) {
|
|
|
|
ret = PTR_ERR(file);
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
|
|
|
|
eventfd = eventfd_ctx_fileget(file);
|
|
|
|
if (IS_ERR(eventfd)) {
|
|
|
|
ret = PTR_ERR(eventfd);
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
|
|
|
|
irqfd->eventfd = eventfd;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Install our own custom wake-up handling so we are notified via
|
|
|
|
* a callback whenever someone signals the underlying eventfd
|
|
|
|
*/
|
|
|
|
init_waitqueue_func_entry(&irqfd->wait, irqfd_wakeup);
|
|
|
|
init_poll_funcptr(&irqfd->pt, irqfd_ptable_queue_proc);
|
|
|
|
|
|
|
|
events = file->f_op->poll(file, &irqfd->pt);
|
|
|
|
|
|
|
|
spin_lock_irq(&kvm->irqfds.lock);
|
|
|
|
list_add_tail(&irqfd->list, &kvm->irqfds.items);
|
|
|
|
spin_unlock_irq(&kvm->irqfds.lock);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Check if there was an event already pending on the eventfd
|
|
|
|
* before we registered, and trigger it as if we didn't miss it.
|
|
|
|
*/
|
|
|
|
if (events & POLLIN)
|
|
|
|
schedule_work(&irqfd->inject);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* do not drop the file until the irqfd is fully initialized, otherwise
|
|
|
|
* we might race against the POLLHUP
|
|
|
|
*/
|
|
|
|
fput(file);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
fail:
|
|
|
|
if (eventfd && !IS_ERR(eventfd))
|
|
|
|
eventfd_ctx_put(eventfd);
|
|
|
|
|
2009-07-28 22:53:24 +07:00
|
|
|
if (!IS_ERR(file))
|
2009-05-20 21:30:49 +07:00
|
|
|
fput(file);
|
|
|
|
|
|
|
|
kfree(irqfd);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
KVM: add ioeventfd support
ioeventfd is a mechanism to register PIO/MMIO regions to trigger an eventfd
signal when written to by a guest. Host userspace can register any
arbitrary IO address with a corresponding eventfd and then pass the eventfd
to a specific end-point of interest for handling.
Normal IO requires a blocking round-trip since the operation may cause
side-effects in the emulated model or may return data to the caller.
Therefore, an IO in KVM traps from the guest to the host, causes a VMX/SVM
"heavy-weight" exit back to userspace, and is ultimately serviced by qemu's
device model synchronously before returning control back to the vcpu.
However, there is a subclass of IO which acts purely as a trigger for
other IO (such as to kick off an out-of-band DMA request, etc). For these
patterns, the synchronous call is particularly expensive since we really
only want to simply get our notification transmitted asychronously and
return as quickly as possible. All the sychronous infrastructure to ensure
proper data-dependencies are met in the normal IO case are just unecessary
overhead for signalling. This adds additional computational load on the
system, as well as latency to the signalling path.
Therefore, we provide a mechanism for registration of an in-kernel trigger
point that allows the VCPU to only require a very brief, lightweight
exit just long enough to signal an eventfd. This also means that any
clients compatible with the eventfd interface (which includes userspace
and kernelspace equally well) can now register to be notified. The end
result should be a more flexible and higher performance notification API
for the backend KVM hypervisor and perhipheral components.
To test this theory, we built a test-harness called "doorbell". This
module has a function called "doorbell_ring()" which simply increments a
counter for each time the doorbell is signaled. It supports signalling
from either an eventfd, or an ioctl().
We then wired up two paths to the doorbell: One via QEMU via a registered
io region and through the doorbell ioctl(). The other is direct via
ioeventfd.
You can download this test harness here:
ftp://ftp.novell.com/dev/ghaskins/doorbell.tar.bz2
The measured results are as follows:
qemu-mmio: 110000 iops, 9.09us rtt
ioeventfd-mmio: 200100 iops, 5.00us rtt
ioeventfd-pio: 367300 iops, 2.72us rtt
I didn't measure qemu-pio, because I have to figure out how to register a
PIO region with qemu's device model, and I got lazy. However, for now we
can extrapolate based on the data from the NULLIO runs of +2.56us for MMIO,
and -350ns for HC, we get:
qemu-pio: 153139 iops, 6.53us rtt
ioeventfd-hc: 412585 iops, 2.37us rtt
these are just for fun, for now, until I can gather more data.
Here is a graph for your convenience:
http://developer.novell.com/wiki/images/7/76/Iofd-chart.png
The conclusion to draw is that we save about 4us by skipping the userspace
hop.
--------------------
Signed-off-by: Gregory Haskins <ghaskins@novell.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
2009-07-08 04:08:49 +07:00
|
|
|
kvm_eventfd_init(struct kvm *kvm)
|
2009-05-20 21:30:49 +07:00
|
|
|
{
|
|
|
|
spin_lock_init(&kvm->irqfds.lock);
|
|
|
|
INIT_LIST_HEAD(&kvm->irqfds.items);
|
KVM: add ioeventfd support
ioeventfd is a mechanism to register PIO/MMIO regions to trigger an eventfd
signal when written to by a guest. Host userspace can register any
arbitrary IO address with a corresponding eventfd and then pass the eventfd
to a specific end-point of interest for handling.
Normal IO requires a blocking round-trip since the operation may cause
side-effects in the emulated model or may return data to the caller.
Therefore, an IO in KVM traps from the guest to the host, causes a VMX/SVM
"heavy-weight" exit back to userspace, and is ultimately serviced by qemu's
device model synchronously before returning control back to the vcpu.
However, there is a subclass of IO which acts purely as a trigger for
other IO (such as to kick off an out-of-band DMA request, etc). For these
patterns, the synchronous call is particularly expensive since we really
only want to simply get our notification transmitted asychronously and
return as quickly as possible. All the sychronous infrastructure to ensure
proper data-dependencies are met in the normal IO case are just unecessary
overhead for signalling. This adds additional computational load on the
system, as well as latency to the signalling path.
Therefore, we provide a mechanism for registration of an in-kernel trigger
point that allows the VCPU to only require a very brief, lightweight
exit just long enough to signal an eventfd. This also means that any
clients compatible with the eventfd interface (which includes userspace
and kernelspace equally well) can now register to be notified. The end
result should be a more flexible and higher performance notification API
for the backend KVM hypervisor and perhipheral components.
To test this theory, we built a test-harness called "doorbell". This
module has a function called "doorbell_ring()" which simply increments a
counter for each time the doorbell is signaled. It supports signalling
from either an eventfd, or an ioctl().
We then wired up two paths to the doorbell: One via QEMU via a registered
io region and through the doorbell ioctl(). The other is direct via
ioeventfd.
You can download this test harness here:
ftp://ftp.novell.com/dev/ghaskins/doorbell.tar.bz2
The measured results are as follows:
qemu-mmio: 110000 iops, 9.09us rtt
ioeventfd-mmio: 200100 iops, 5.00us rtt
ioeventfd-pio: 367300 iops, 2.72us rtt
I didn't measure qemu-pio, because I have to figure out how to register a
PIO region with qemu's device model, and I got lazy. However, for now we
can extrapolate based on the data from the NULLIO runs of +2.56us for MMIO,
and -350ns for HC, we get:
qemu-pio: 153139 iops, 6.53us rtt
ioeventfd-hc: 412585 iops, 2.37us rtt
these are just for fun, for now, until I can gather more data.
Here is a graph for your convenience:
http://developer.novell.com/wiki/images/7/76/Iofd-chart.png
The conclusion to draw is that we save about 4us by skipping the userspace
hop.
--------------------
Signed-off-by: Gregory Haskins <ghaskins@novell.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
2009-07-08 04:08:49 +07:00
|
|
|
INIT_LIST_HEAD(&kvm->ioeventfds);
|
2009-05-20 21:30:49 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* shutdown any irqfd's that match fd+gsi
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
kvm_irqfd_deassign(struct kvm *kvm, int fd, int gsi)
|
|
|
|
{
|
|
|
|
struct _irqfd *irqfd, *tmp;
|
|
|
|
struct eventfd_ctx *eventfd;
|
|
|
|
|
|
|
|
eventfd = eventfd_ctx_fdget(fd);
|
|
|
|
if (IS_ERR(eventfd))
|
|
|
|
return PTR_ERR(eventfd);
|
|
|
|
|
|
|
|
spin_lock_irq(&kvm->irqfds.lock);
|
|
|
|
|
|
|
|
list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list) {
|
|
|
|
if (irqfd->eventfd == eventfd && irqfd->gsi == gsi)
|
|
|
|
irqfd_deactivate(irqfd);
|
|
|
|
}
|
|
|
|
|
|
|
|
spin_unlock_irq(&kvm->irqfds.lock);
|
|
|
|
eventfd_ctx_put(eventfd);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Block until we know all outstanding shutdown jobs have completed
|
|
|
|
* so that we guarantee there will not be any more interrupts on this
|
|
|
|
* gsi once this deassign function returns.
|
|
|
|
*/
|
|
|
|
flush_workqueue(irqfd_cleanup_wq);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int
|
|
|
|
kvm_irqfd(struct kvm *kvm, int fd, int gsi, int flags)
|
|
|
|
{
|
|
|
|
if (flags & KVM_IRQFD_FLAG_DEASSIGN)
|
|
|
|
return kvm_irqfd_deassign(kvm, fd, gsi);
|
|
|
|
|
|
|
|
return kvm_irqfd_assign(kvm, fd, gsi);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This function is called as the kvm VM fd is being released. Shutdown all
|
|
|
|
* irqfds that still remain open
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
kvm_irqfd_release(struct kvm *kvm)
|
|
|
|
{
|
|
|
|
struct _irqfd *irqfd, *tmp;
|
|
|
|
|
|
|
|
spin_lock_irq(&kvm->irqfds.lock);
|
|
|
|
|
|
|
|
list_for_each_entry_safe(irqfd, tmp, &kvm->irqfds.items, list)
|
|
|
|
irqfd_deactivate(irqfd);
|
|
|
|
|
|
|
|
spin_unlock_irq(&kvm->irqfds.lock);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Block until we know all outstanding shutdown jobs have completed
|
|
|
|
* since we do not take a kvm* reference.
|
|
|
|
*/
|
|
|
|
flush_workqueue(irqfd_cleanup_wq);
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* create a host-wide workqueue for issuing deferred shutdown requests
|
|
|
|
* aggregated from all vm* instances. We need our own isolated single-thread
|
|
|
|
* queue to prevent deadlock against flushing the normal work-queue.
|
|
|
|
*/
|
|
|
|
static int __init irqfd_module_init(void)
|
|
|
|
{
|
|
|
|
irqfd_cleanup_wq = create_singlethread_workqueue("kvm-irqfd-cleanup");
|
|
|
|
if (!irqfd_cleanup_wq)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void __exit irqfd_module_exit(void)
|
|
|
|
{
|
|
|
|
destroy_workqueue(irqfd_cleanup_wq);
|
|
|
|
}
|
|
|
|
|
|
|
|
module_init(irqfd_module_init);
|
|
|
|
module_exit(irqfd_module_exit);
|
KVM: add ioeventfd support
ioeventfd is a mechanism to register PIO/MMIO regions to trigger an eventfd
signal when written to by a guest. Host userspace can register any
arbitrary IO address with a corresponding eventfd and then pass the eventfd
to a specific end-point of interest for handling.
Normal IO requires a blocking round-trip since the operation may cause
side-effects in the emulated model or may return data to the caller.
Therefore, an IO in KVM traps from the guest to the host, causes a VMX/SVM
"heavy-weight" exit back to userspace, and is ultimately serviced by qemu's
device model synchronously before returning control back to the vcpu.
However, there is a subclass of IO which acts purely as a trigger for
other IO (such as to kick off an out-of-band DMA request, etc). For these
patterns, the synchronous call is particularly expensive since we really
only want to simply get our notification transmitted asychronously and
return as quickly as possible. All the sychronous infrastructure to ensure
proper data-dependencies are met in the normal IO case are just unecessary
overhead for signalling. This adds additional computational load on the
system, as well as latency to the signalling path.
Therefore, we provide a mechanism for registration of an in-kernel trigger
point that allows the VCPU to only require a very brief, lightweight
exit just long enough to signal an eventfd. This also means that any
clients compatible with the eventfd interface (which includes userspace
and kernelspace equally well) can now register to be notified. The end
result should be a more flexible and higher performance notification API
for the backend KVM hypervisor and perhipheral components.
To test this theory, we built a test-harness called "doorbell". This
module has a function called "doorbell_ring()" which simply increments a
counter for each time the doorbell is signaled. It supports signalling
from either an eventfd, or an ioctl().
We then wired up two paths to the doorbell: One via QEMU via a registered
io region and through the doorbell ioctl(). The other is direct via
ioeventfd.
You can download this test harness here:
ftp://ftp.novell.com/dev/ghaskins/doorbell.tar.bz2
The measured results are as follows:
qemu-mmio: 110000 iops, 9.09us rtt
ioeventfd-mmio: 200100 iops, 5.00us rtt
ioeventfd-pio: 367300 iops, 2.72us rtt
I didn't measure qemu-pio, because I have to figure out how to register a
PIO region with qemu's device model, and I got lazy. However, for now we
can extrapolate based on the data from the NULLIO runs of +2.56us for MMIO,
and -350ns for HC, we get:
qemu-pio: 153139 iops, 6.53us rtt
ioeventfd-hc: 412585 iops, 2.37us rtt
these are just for fun, for now, until I can gather more data.
Here is a graph for your convenience:
http://developer.novell.com/wiki/images/7/76/Iofd-chart.png
The conclusion to draw is that we save about 4us by skipping the userspace
hop.
--------------------
Signed-off-by: Gregory Haskins <ghaskins@novell.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
2009-07-08 04:08:49 +07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* --------------------------------------------------------------------
|
|
|
|
* ioeventfd: translate a PIO/MMIO memory write to an eventfd signal.
|
|
|
|
*
|
|
|
|
* userspace can register a PIO/MMIO address with an eventfd for receiving
|
|
|
|
* notification when the memory has been touched.
|
|
|
|
* --------------------------------------------------------------------
|
|
|
|
*/
|
|
|
|
|
|
|
|
struct _ioeventfd {
|
|
|
|
struct list_head list;
|
|
|
|
u64 addr;
|
|
|
|
int length;
|
|
|
|
struct eventfd_ctx *eventfd;
|
|
|
|
u64 datamatch;
|
|
|
|
struct kvm_io_device dev;
|
|
|
|
bool wildcard;
|
|
|
|
};
|
|
|
|
|
|
|
|
static inline struct _ioeventfd *
|
|
|
|
to_ioeventfd(struct kvm_io_device *dev)
|
|
|
|
{
|
|
|
|
return container_of(dev, struct _ioeventfd, dev);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
ioeventfd_release(struct _ioeventfd *p)
|
|
|
|
{
|
|
|
|
eventfd_ctx_put(p->eventfd);
|
|
|
|
list_del(&p->list);
|
|
|
|
kfree(p);
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool
|
|
|
|
ioeventfd_in_range(struct _ioeventfd *p, gpa_t addr, int len, const void *val)
|
|
|
|
{
|
|
|
|
u64 _val;
|
|
|
|
|
|
|
|
if (!(addr == p->addr && len == p->length))
|
|
|
|
/* address-range must be precise for a hit */
|
|
|
|
return false;
|
|
|
|
|
|
|
|
if (p->wildcard)
|
|
|
|
/* all else equal, wildcard is always a hit */
|
|
|
|
return true;
|
|
|
|
|
|
|
|
/* otherwise, we have to actually compare the data */
|
|
|
|
|
|
|
|
BUG_ON(!IS_ALIGNED((unsigned long)val, len));
|
|
|
|
|
|
|
|
switch (len) {
|
|
|
|
case 1:
|
|
|
|
_val = *(u8 *)val;
|
|
|
|
break;
|
|
|
|
case 2:
|
|
|
|
_val = *(u16 *)val;
|
|
|
|
break;
|
|
|
|
case 4:
|
|
|
|
_val = *(u32 *)val;
|
|
|
|
break;
|
|
|
|
case 8:
|
|
|
|
_val = *(u64 *)val;
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
return _val == p->datamatch ? true : false;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* MMIO/PIO writes trigger an event if the addr/val match */
|
|
|
|
static int
|
|
|
|
ioeventfd_write(struct kvm_io_device *this, gpa_t addr, int len,
|
|
|
|
const void *val)
|
|
|
|
{
|
|
|
|
struct _ioeventfd *p = to_ioeventfd(this);
|
|
|
|
|
|
|
|
if (!ioeventfd_in_range(p, addr, len, val))
|
|
|
|
return -EOPNOTSUPP;
|
|
|
|
|
|
|
|
eventfd_signal(p->eventfd, 1);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This function is called as KVM is completely shutting down. We do not
|
|
|
|
* need to worry about locking just nuke anything we have as quickly as possible
|
|
|
|
*/
|
|
|
|
static void
|
|
|
|
ioeventfd_destructor(struct kvm_io_device *this)
|
|
|
|
{
|
|
|
|
struct _ioeventfd *p = to_ioeventfd(this);
|
|
|
|
|
|
|
|
ioeventfd_release(p);
|
|
|
|
}
|
|
|
|
|
|
|
|
static const struct kvm_io_device_ops ioeventfd_ops = {
|
|
|
|
.write = ioeventfd_write,
|
|
|
|
.destructor = ioeventfd_destructor,
|
|
|
|
};
|
|
|
|
|
|
|
|
/* assumes kvm->slots_lock held */
|
|
|
|
static bool
|
|
|
|
ioeventfd_check_collision(struct kvm *kvm, struct _ioeventfd *p)
|
|
|
|
{
|
|
|
|
struct _ioeventfd *_p;
|
|
|
|
|
|
|
|
list_for_each_entry(_p, &kvm->ioeventfds, list)
|
|
|
|
if (_p->addr == p->addr && _p->length == p->length &&
|
|
|
|
(_p->wildcard || p->wildcard ||
|
|
|
|
_p->datamatch == p->datamatch))
|
|
|
|
return true;
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
kvm_assign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
|
|
|
|
{
|
|
|
|
int pio = args->flags & KVM_IOEVENTFD_FLAG_PIO;
|
|
|
|
struct kvm_io_bus *bus = pio ? &kvm->pio_bus : &kvm->mmio_bus;
|
|
|
|
struct _ioeventfd *p;
|
|
|
|
struct eventfd_ctx *eventfd;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
/* must be natural-word sized */
|
|
|
|
switch (args->len) {
|
|
|
|
case 1:
|
|
|
|
case 2:
|
|
|
|
case 4:
|
|
|
|
case 8:
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* check for range overflow */
|
|
|
|
if (args->addr + args->len < args->addr)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
/* check for extra flags that we don't understand */
|
|
|
|
if (args->flags & ~KVM_IOEVENTFD_VALID_FLAG_MASK)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
eventfd = eventfd_ctx_fdget(args->fd);
|
|
|
|
if (IS_ERR(eventfd))
|
|
|
|
return PTR_ERR(eventfd);
|
|
|
|
|
|
|
|
p = kzalloc(sizeof(*p), GFP_KERNEL);
|
|
|
|
if (!p) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
|
|
|
|
INIT_LIST_HEAD(&p->list);
|
|
|
|
p->addr = args->addr;
|
|
|
|
p->length = args->len;
|
|
|
|
p->eventfd = eventfd;
|
|
|
|
|
|
|
|
/* The datamatch feature is optional, otherwise this is a wildcard */
|
|
|
|
if (args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH)
|
|
|
|
p->datamatch = args->datamatch;
|
|
|
|
else
|
|
|
|
p->wildcard = true;
|
|
|
|
|
|
|
|
down_write(&kvm->slots_lock);
|
|
|
|
|
|
|
|
/* Verify that there isnt a match already */
|
|
|
|
if (ioeventfd_check_collision(kvm, p)) {
|
|
|
|
ret = -EEXIST;
|
|
|
|
goto unlock_fail;
|
|
|
|
}
|
|
|
|
|
|
|
|
kvm_iodevice_init(&p->dev, &ioeventfd_ops);
|
|
|
|
|
|
|
|
ret = __kvm_io_bus_register_dev(bus, &p->dev);
|
|
|
|
if (ret < 0)
|
|
|
|
goto unlock_fail;
|
|
|
|
|
|
|
|
list_add_tail(&p->list, &kvm->ioeventfds);
|
|
|
|
|
|
|
|
up_write(&kvm->slots_lock);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
unlock_fail:
|
|
|
|
up_write(&kvm->slots_lock);
|
|
|
|
|
|
|
|
fail:
|
|
|
|
kfree(p);
|
|
|
|
eventfd_ctx_put(eventfd);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
kvm_deassign_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
|
|
|
|
{
|
|
|
|
int pio = args->flags & KVM_IOEVENTFD_FLAG_PIO;
|
|
|
|
struct kvm_io_bus *bus = pio ? &kvm->pio_bus : &kvm->mmio_bus;
|
|
|
|
struct _ioeventfd *p, *tmp;
|
|
|
|
struct eventfd_ctx *eventfd;
|
|
|
|
int ret = -ENOENT;
|
|
|
|
|
|
|
|
eventfd = eventfd_ctx_fdget(args->fd);
|
|
|
|
if (IS_ERR(eventfd))
|
|
|
|
return PTR_ERR(eventfd);
|
|
|
|
|
|
|
|
down_write(&kvm->slots_lock);
|
|
|
|
|
|
|
|
list_for_each_entry_safe(p, tmp, &kvm->ioeventfds, list) {
|
|
|
|
bool wildcard = !(args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH);
|
|
|
|
|
|
|
|
if (p->eventfd != eventfd ||
|
|
|
|
p->addr != args->addr ||
|
|
|
|
p->length != args->len ||
|
|
|
|
p->wildcard != wildcard)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (!p->wildcard && p->datamatch != args->datamatch)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
__kvm_io_bus_unregister_dev(bus, &p->dev);
|
|
|
|
ioeventfd_release(p);
|
|
|
|
ret = 0;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
up_write(&kvm->slots_lock);
|
|
|
|
|
|
|
|
eventfd_ctx_put(eventfd);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
int
|
|
|
|
kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
|
|
|
|
{
|
|
|
|
if (args->flags & KVM_IOEVENTFD_FLAG_DEASSIGN)
|
|
|
|
return kvm_deassign_ioeventfd(kvm, args);
|
|
|
|
|
|
|
|
return kvm_assign_ioeventfd(kvm, args);
|
|
|
|
}
|